AUTOMATIC DETECTION OF THE NUMBER OF LANES INTO WHICH A ROAD IS DIVIDED
This invention concerns a computer-implemented method for determining a number of lanes on a road. The method comprises receiving an image, the image being a photographic image of the road or an image derived from the photographic image of the road, carrying out an analysis of the image to identify lane dividers of the road and determining a number of lanes into which the road is divided from the identification of the lane dividers. The method may comprise determining a confidence level for the determined value for the number of lanes. Map data of a plurality of roads may be generated using the value for the number of lanes determined from the computer-implemented method.
This invention concerns improvements in or relating to automatic detection of the number of lanes into which a road is divided. The invention has particular application to the automatic detection of the number of lanes into which a road is divided using photographic images of the road or images derived from the photographic images and the generation of map data therefrom.
BACKGROUND OF THE INVENTIONTo generate map data for use by navigation devices, such as Portable Navigation Devices (PNDs), it is desirable to determine the number of lanes into which roads are divided (hereinafter referred to as “lane count”). The navigation device is then able to provide accurate information and instructions concerning a route being driven, such as a display of the lane count. The determination of the lane count can be carried out manually. However, such manual identification is slow and involves significant labour costs.
To reduce the time and amount of manual labour required in determining lane count, a method for automatically identifying the lane count is used. One such method analyses data, such as position data, collected from Tom Tom navigation devices in vehicles as the vehicles travel the road (hereinafter referred to as “probe data”). Below is a table of the results of such a method applied to roads where the number of lanes is known.
The table shows the number of roads automatically identified as having a lane count of 2, 3, 4 or 5 from the probe data, denoted “out”, against a reference value, denoted “Ref”. As can be seen, for the 3907 roads having two lanes, the automatic analysis of the probe data only identifies 3283 roads as having two lanes; the automatic analysis misidentifying other two lane roads as having more lanes. Similar results are obtained for roads having other lane counts. The automatic identification of lane count from probe data is particularly poor for roads with five lanes.
It is desirable to improve the accuracy of the automated lane counting.
SUMMARY OF INVENTIONAccording to a first aspect of the invention there is provided a computer implemented method for determining a number of lanes on a road comprising receiving an image, the image being a photographic image of the road or an image derived from the photographic image of the road, carrying out an analysis of the image to identify lane dividers of the road and determining a number of lanes into which the road is divided from the identification of the lane dividers.
It has been found that methods in accordance with the invention can identify the number of lanes into which a road is divided more accurately than the automatic analysis of probe data alone.
In one embodiment, the analysis is a spectral analysis. The spectral analysis may comprise searching portions of the image for one or more spectral signatures characteristic of a lane divider. It has been found that a lane divider has a readily identifiable spectral signature allowing lane dividers to be identified from photographic images or images derived from photographic images. The spectral analysis may comprise generating an image frequency spectrum by carrying out a Fourier transform of the image and comparing at least a portion of the image frequency spectrum with one or more spectral signatures characteristic of a lane divider. The comparison may comprise searching for a set of characteristic local maxima within the image frequency spectrum. The comparison may comprise comparing a ratio of frequencies of the local maxima, comparing a magnitude of the local maxima, determining whether one or more of the local maxima are above a predetermined threshold, comparing average values of portions (and possibly the whole) of the spectrum to a magnitude of the local maxima. If one or more or all of these comparisons are found that are characteristic of spectral signatures of a lane divider, a portion of the image is identified as containing a lane divider.
The method may comprise identifying road boundaries in the image and confining a search for lane dividers to a portion of the image within the identified road boundaries. The search for lane dividers may comprise searching for one or more spectral signatures characteristic of a lane divider within the identified road boundaries. Confining the search for lane dividers to be within the road boundaries reduces the amount of processing required and reduces the chance of portions of the image being incorrectly identified as lane dividers.
The method may comprise filtering the image before the spectral analysis to confine the search for lane dividers to one or more portions of the road that are substantially straight in a first (ordinate) direction. For example, the image may be an LRI image localised along a centreline of the road and the search may be confined to one or more portions of the image between two road boundaries that, for a plurality of possible road boundaries, have the two smallest variances in a second (abscissa) direction over minimum length in the first direction.
The analysis may comprise dividing the image into columns extending in the first ordinate direction and comparing an image spectrum of each column to a signature characteristic of a lane divider. Columns whose image spectrum comprises elements the same or substantially similar to the signature characteristic of a lane divider are identified as comprising a lane divider.
The analysis of the image may comprise determining a width of the lanes from the elements of the image identified lane dividers. The method may comprise rejecting as a lane a space between two identified lane dividers if the space has a width above a predetermined upper width threshold or below a predetermined lower width threshold. The predetermined upper width threshold may be between 3 to 5 metres, and preferably 4 metres, and the predetermined lower width threshold may be between 0.5 to 2 metres, and preferably 1.5 metres.
The method may comprise determining a confidence level for the determined number of lanes (lane count). The method may comprise rejecting the lane count if the determined confidence level is below a predetermined confidence threshold. In this way, an indication is provided of any poor results allowing such poor results to be removed when utilizing the data.
The method may comprise determined the lane count from features derived from the image and features derived from received probe data.
The method may comprise dividing the image into a number of sub-images and carrying out the analysis on the sub-images. This may increase the reliability of the lane count as the lane count of a road may vary over large distances.
According to a second aspect of the invention there is provided a processing device arranged to carry out the method of the first aspect of the invention.
According to a third aspect of the invention there is provided a data carrier having stored thereon instructions, which, when executed by a processor cause the processor to execute a method according to the first aspect of the invention.
According to a fourth aspect of the invention there is provided a method of generating map data of a plurality of roads comprising determining a number of lanes on one or more of the roads using the method of the first aspect of the invention and generating map data identifying the number of lanes.
There now follows, by way of example only, a detailed description of embodiments of the present invention with reference to the accompanying drawings, in which:
Referring to
Now referring to
Each sub-image is then processed in block 108 to identify lane dividers in the sub-image and the total width of the road.
In block 109, lane dividers are identified by first generating road boundaries through conventional morphological operations, filtering and segmentation. Road boundaries are the two possible road boundaries that, for a plurality of possible road boundaries, have the two smallest variances in the abscissa direction over a minimum distance in the ordinate direction, Z. In this embodiment, the minimal distance is 100 pixels of the sub-image. For example, for candidate road boundaries may be identified having variances over 100 pixels of 0.98, 0.02. 0.032 and 0.23. The candidate road boundaries having the two smallest variances of 0.02 and 0.032 are identified as the road boundaries within which the search for lane dividers is confined.
The search for lane dividers is confined to the one or more portions of the sub-image that are within the road boundaries. A spectral analysis is then carried out on these portions of the sub-image to identify lane dividers. In this embodiment, the sub-image is divided along the abscissa direction into columns that extend in the ordinate direction. A Fourier transform of each column is then compared to spectral signatures characteristic of a lane divider. Columns whose image spectrum comprises elements the same or substantially similar to the spectral signatures characteristic of a lane divider are identified as comprising a lane divider. A characteristic frequency spectrum of a lane divider is shown in
- 1. That the first maximum is located above a minimum frequency, in this embodiment 5.
- 2. That the ratio of the local maxima, in this embodiment the first local maxima to the second local maxima, is within a predefined range, in this embodiment between 1.8 and 2.2.
3. That the magnitude of the at least one of the local maxima, in this embodiment the first local maxima, is above a predetermined threshold, in this embodiment above 0.35.
4. That the ratio of the values of the local maxima, in this embodiment the first local maxima to the second local maxima, is within a predetermined range, in this embodiment, between 1.05 and 2.5.
5. The average value (in this embodiment, the median value) of the frequencies above the second local maxima should be less than the magnitude of the second local maxima.
6. The average value (in this embodiment, the median value) of the frequencies below the first local maxima should be less than a predetermined proportion of the magnitude of the first local maxima, in this embodiment less than half of the magnitude of the first local maxima.
Columns comprising these spectral characteristics are identified as comprising a lane divider.
The results of the analysis of block 109 are passed to block 110 wherein a width of the lanes is determined from the identified lane dividers. In this embodiment, the width of the road is determined in pixels and then using data supplied with the LRI 13 image, the pixels are converted in to metres.
The features determined from the image are then combined with features determined from the probe data, an example of such data being shown in
y=(Ax1+Bx2+Cx3+Dx4+Ex5+Fx6+Gx7+Hx8+Ix9+Jx10+Kx11+Lx12+Mx13)×N+O, (1)
where:
x1 is the lane count obtained from the probe data
x2 is the road width calculated from the probe data
x3 is image lane count obtained from the probe data×average distance between histogram peaks of the probe data
x4 is peak magnitude of the histogram of the probe data. This peak is measured as the distance between the peak and the valley near to this peak.
If there are several peaks, an average value for all peaks is determined.
x5 is lane count from the probe data
x6 is a width of the histogram for the probe data
x7 is the standard deviation for the probe data
x8 is the width of the probe data that contains 68 percent of the samples
x9 is the width of the probe data that contains 95 percent of the samples
x10 is the distance to the left road border from the road centreline as determined from the probe data
x11 is the distance to the right road border from the road centreline as determined from the probe data
x12 is the number of lane dividers determined from the image
x13 is the total width (road width) between the dividers as determined from the image
The values of constants A to O have been determined from an optimization procedure that minimizes the error between a computed lane count and a known lane count for a particular set of “training” roads. However, it will be understood that other values may be used, such as more or less accurate values. However, in this embodiment, the following values are used:
A=−0.081177398
B=0.191450131
C=0.015670102
D=−1.461469312
E=7.16335×10−5
F=0.004279486
G=0.474283555
H=−0.077535198
I=0.372897898
J=0.400163453
K=−0.360302749
L=−0.012281159
M=0.100657595
N=1.3613
O=−0.968
The value of y may be between 1.5 and 5.5 and the actual lane count for a road is determined by determining the nearest neighbouring integer 1, 2, 3, 4 of 5 for the value y.
An example of a distribution of the values of y determined for multiple road segments is shown in
A confidence level function has been determined from which a confidence level can be attributed to any given value y. This confidence level function is given by:
where,
N(i, σi) is a normal distribution with a standard deviation as follows:
It has been found that a method in accordance with this embodiment of the invention can identify the number of lanes into which a road is divided more accurately than the automatic analysis of probe data alone. This is illustrated by the following table of a reference lane count (for the same set of roads used for table 1) versus lane count extracted from probe data and LRI images using the above-described method.
As can be seen, the number of correct lane counts has increased relative to using the probe data alone, the results of which are shown in table 1. Adding the results of lane count extracted from the LRI 13 image improved detection level by 13.5% (from 65.4% to 78.9%)
The results of the method can be used for the generation of map data, which may be used in navigation devices. In particular, the calculated confidence level may be used to decide whether the lane count automatically determined by the method should be used, for example in map data, or whether a further source of lane count should be consulted to determine lane count for a particular road.
It will be understood that modifications and alterations can be made to the above described embodiment without departing from the invention as defined in the claims.
Furthermore, the value, y, may be determined using more or fewer arguments with corresponding adjustments to the constants.
The lane count may be determined without determination of a confidence level.
Claims
1. A computer-implemented method for determining a number of lanes on a road comprising receiving an image, the image being a photographic image of the road or an image derived from the photographic image of the road, carrying out an analysis of the image to identify lane dividers of the road and determining a number of lanes into which the road is divided from the identification of the lane dividers.
2. The computer-implemented method according to claim 1, wherein the analysis comprises searching portions of the image for one or more spectral signatures characteristic of a lane divider.
3. The computer-implemented method according to claim 2, wherein the spectral analysis comprises generating an image frequency spectrum by carrying out a Fourier transform of the image and comparing the image frequency spectrum with spectral signatures characteristic of the lane divider.
4. The computer-implemented method according to claim 3, wherein the comparing step comprises searching for a set of characteristic local maxima within the image frequency spectrum.
5. The computer-implemented method according to claim 1, comprising identifying road boundaries in the image and confining a search for lane dividers to a portion of the image within the identified road boundaries.
6. The computer-implemented method according to claim 1, wherein the analysis comprises dividing the image into columns extending in the first, ordinate direction, comparing an image spectrum of each column to a signature characteristic of a lane divider and identifying columns whose image spectrum comprises elements the same or substantially similar to the signature characteristic of a lane divider as comprising a lane divider.
7. The computer-implemented method according to claim 1, wherein the analysis of the image comprises determining a width of the lanes from the elements of the image identified as lane dividers.
8. The computer-implemented method according to claim 7, comprising rejecting as a lane a space between two identified lane dividers if the space has a width above a predetermined upper width threshold or below a predetermined lower width threshold.
9. The computer-implemented method according to claim 1, comprising receiving probe data and determining the lane count from attributes of the image and attributes of the probe data.
10. The computer-implemented method according to claim 1, comprising determining a confidence level for the determined number of lanes.
11. The computer-implemented method according to claim 10, comprising rejecting the number of lanes if the determined confidence level is below a predetermined confidence threshold.
12. (canceled)
13. A non-transitory data carrier having stored thereon instructions, which, when executed by a processor cause the processor to execute a method for determining a number of lanes on a road comprising receiving an image, the image being a photographic image of the road or an image derived from the photographic image of the road, carrying out an analysis of the image to identify lane dividers of the road and determining a number of lanes into which the road is divided from the identification of the lane dividers.
14. (canceled)
15. The computer-implemented method according to claim 1, further comprising generating and storing map data for the road identifying the determined number of lanes.
Type: Application
Filed: Sep 16, 2010
Publication Date: Jul 18, 2013
Patent Grant number: 9355321
Inventor: Artur Wujcicki (Lodz)
Application Number: 13/824,600
International Classification: G06K 9/00 (20060101);